AI Paper Atlas

Research map / Optimization / Optimizers

Learning to Mutate with Hypergradient Guided Population

Zhiqiang Tao, Yaliang Li, Bolin Ding, Ce Zhang, Jingren Zhou, Yun Fu

NeurIPS 2020 · Optimization › Optimizers · approach: Differentiating through inner loops · 5 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add learnable teacher-network mutation (vs random perturbation) (ablation)HPM · CIFAR10validation loss: 0.5724 -> 0.5636Improved
add learnable teacher-network mutation (vs random perturbation) (ablation)HPM · PTBvalidation perplexity: 73.18 -> 70.49Improved
remove mutation, use extra hypergradient step (ablation)HPM · CIFAR10validation loss: 0.5636 -> 0.6139Hurt
remove mutation, use extra hypergradient step (ablation)HPM · PTBvalidation perplexity: 70.49 -> 75.24Hurt
freeze hyperparameters in hypertraining, only mutate (ablation)HPM · CIFAR10validation loss: 0.5636 -> 0.5944Hurt
freeze hyperparameters in hypertraining, only mutate (ablation)HPM · PTBvalidation perplexity: 70.49 -> 73.76Hurt
combine STN with random search instead of PBT exploit/explore (ablation)STN · CIFAR10validation loss: 0.5892 -> 0.5817No clear effect
replace attention teacher with MLP teacher (ablation)HPM · CIFAR10validation loss: 0.5636 -> 0.5696No clear effect
replace attention teacher with MLP teacher (ablation)HPM · PTBvalidation perplexity: 70.49 -> 70.91No clear effect
add hypergradient-guided learnable mutation on top of PBTPBT · CIFAR10validation loss: 0.6253 -> 0.5636Improved
add hypergradient-guided learnable mutation on top of STNSTN · PTBvalidation perplexity: 71.49 -> 70.49Improved
schedule hyperparameters instead of fixed configurationhyperparameter schedule methods · CIFAR10validation loss: 0.6360 -> 0.5636Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy